An English board is usually optimized for a global launch story: English landing page, USD pricing, docs in English, discovery via Product Hunt / HN / Twitter. That is rational if your reader already lives in that stack. It is a poor map if the reader’s real question is “which of these will I still be able to open, pay for, and finish a job with from here — this afternoon?”

A bilingual directory is not “translate every blurb into two languages.” It is a different decision tree for what gets a row, what gets a warning, and what gets left out.

What English catalogs systematically underspecify

Three gaps show up again and again when you compare a US-centric list to how Chinese developers and creators actually move:

1. Access and account friction are treated as edge cases.
   An English catalog can list a tool as “free to try” when the signup only completes with a phone number, card, or region the reader does not have. The listing is not lying in English-market terms. It is incomplete for a large share of practitioners who still need the same category of tool.

2. Docs language is invisible in the card.
   Many “global” tools ship excellent English docs and a thin or machine-translated CN page — or none. For install, auth, and error strings, that gap is the product. A directory that only stores the English docs URL sends CN readers into a dead end they could have avoided with one extra field: docs language / CN mirror / community guide.

3. Local alternatives are filed as “regional” instead of “same job.”
   SiliconFlow next to Together, ModelScope next to Hugging Face, CodeBuddy next to Cursor — these are often the same task class with different rails. English boards either omit the CN option or bury it under a geography tag. Readers comparing “coding agent” or “model API gateway” need the job match first, nationality second.

None of that means English directories are bad. It means their default reader is not the only reader who needs an AI tools map.

What Chinese readers actually click through for

From the other side, the click path is blunt:

• Does the official site open cleanly? Flaky, heavily gated, or VPN-only homepages lose the click even when Twitter loves the tool.
• Is there a CN-readable path to first success? Official CN docs, a maintained mirror note, or a clear “English UI / CN payment” split beats a pretty English screenshot.
• Can I pay or stay free without a surprise? Local rails beat a USD badge that is not actionable.
• Is there a peer tool for the same job with less friction? People will take a slightly weaker model API if the stronger one cannot finish signup. Hiding the peer tool is ideology, not navigation.

Those clicks look “local.” They are task completion clicks. Language and region are just the constraints.

The bilingual tradeoff (and what to refuse)

Running a Chinese-first board that still lists English-named tools means accepting mixed UI chrome, dated access notes you must re-check, and looking “behind” on hype weeks when you refuse Product Hunt order. The gain is comparison shopping that matches real jobs.

The refusal that matters most: do not treat translation as curation. A machine-translated English directory is still an English decision tree with worse prose. Curation means choosing which friction to surface.

WarpNav is one example of that tree: Chinese-maintained navigation and guides, English tool names left intact, shelves for model apps, coding assistants, and gateways that put global and CN-accessible options in the same job buckets. Not “most logos from a launch feed” — a shorter path from “I need this job done” to “here is an official door that might actually open.”

A practical test before you trust any “AI tools” board

1. Does each card state who the listing is for (language, region, account assumptions) — or only the vendor’s English slogan?
2. Are peer tools for the same job visible in one shelf, or siloed as “International” vs “China”?
3. When docs or pricing are English-only / CN-only, is that labeled?
4. If hyped tools failed basic access from your network, did the board still rank them by Twitter heat?

Slogan / silo / unlabeled / heat means you have a launch feed with favicons — not a map for bilingual work.

English catalogs are strong at naming what launched. Bilingual catalogs earn their keep when they name what you can finish.

I wanted a narrower job.

The job

When OpenAI-side Codex posts announce a usage-limit reset — “before 10am PST,” “by EOD,” “all propagated” — remote teams still argue about three different local conversions of the same sentence. Separately, people mash three different mechanisms under one word: your rolling windows, a public/global-style reset, and a banked reset card.

The job I cared about was not “show melonape’s remaining percent.” It was: keep one shared, source-linked board for public reset signals, with honest labels for what is confirmed vs estimated, and a ping when something new lands — including when usage-limit rules change, not only when a countdown hits zero.

What I refused to build

• No menu bar. Menu bars are great for your quota. They are the wrong shape for a team staring at the same X screenshot.
• No CLI that scrapes your account. Personal limits belong in Codex. A public tracker that pretends to know your bar would be lying.
• No “predict the next reset from history.” An archive is useful. A fake schedule is not. History here is evidence of past signals, not a promise of the next day.

That left a webpage on purpose.

What shipped

Codex Reset Watch is a small unofficial site (WarpNav) that tracks confirmed and upcoming Codex usage-limit resets with:

• a live UTC countdown (so the room shares one clock before anyone converts to local),
• original Tibo source posts next to the time claim,
• timezone notes for messy labels like PST / EOD,
• a verified history that separates confirmation timestamps from “we observed it on one account,”
• browser alerts when a new reset is confirmed on the site — and when Codex usage-limit rules change.

No signup wall for opening the board. It does not log into ChatGPT or Codex. It does not speak for OpenAI.

Why “one thing” still has two alert types

Resets are the obvious event. Rule changes are quieter and often more expensive: a Plus 5-hour window coming back, a plan treated differently, wording that shifts who is eligible. If you only alert on “tokens refilled,” you miss the week where the policy moved under people’s feet.

So the product stays narrow in surface (one page, public signals) and slightly wider in signal types (landing events + rule changes). That felt more honest than bolting on token graphs I cannot verify from the outside.

Who it is for

Useful if you coordinate with people in other timezones and you already know how to check your own /status. Less useful if you only want a personal remaining-percent widget — use Codex for that.

I am not trying to win the quota-monitor category. I am trying to make the public clock boring: one UTC reference, one source link, one push when the board updates.

If that is the itch you have, the page is live. If your itch is personal telemetry, skip it — that is a different product, and the ecosystem already has several.